An MCMC Approach to Estimating Demand for Differentiated Durable Products Using Aggregate Data

• Activity Number: 398
• Type: Topic Contributed
• Date/Time: Tuesday, July 31, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Marketing
• Author(s): Yutec Sun*+ and Masakazu Ishihara
• Companies: University of Toronto and New York University
• Keywords: Markov Chain Monte Carlo ; Laplace ; GMM ; Bayes ; Dynamic programming

Abstract

This paper proposes a new approach to estimating structural model of dynamic demand for differentiated durable products using product-level data. The conventional GMM estimator that embeds Rust (1987)'s nested fixed point algorithm has been widely used in the literature. I transform the GMM estimator into a Laplace type estimator (LTE) developed by Chernozhukov and Hong (2003), and use Imai, Jain, and Ching (2009)'s Markov Chain Monte Carlo (MCMC) algorithm to compute the LTE. The proposed approach has two main goals. First, it aims to reduce the computational burden involved with the hierarchical structure of the conventional GMM approach. In doing so, the proposed algorithm does not sacrifice the rich structure of unobserved consumer heterogeneity. Secondly, the proposed LTE avoids making stronger distributional assumptions than the GMM approach, therefore reducing the risk of misspecification bias. This contrasts with alternative approaches such as simulated maximum likelihood, control function, and Bayesian methods that require additional assumptions on either conditional or joint distribution between unobserved common demand shocks and endogenous variables.